| Laser scanning confocal microscopy ( LSCM ) 3D imaging plays an important role in biomedicine, especially in 3D microstructural study. It can be applied in the study of cell morphologies and locations, 3D reconstruction, dynamic changes, etc. It can also be the practical research means for the quantitative measurement and analysis of fluorescence and images. Combined with some other relative biomedical techniques, it has been widely applied in morphology, physiology, pathology, immunology, genetics, and so on.In order to put the laser scanning confocal microscopy techniques into full and efficient use, it is necessary to develop extensively and intensively image processing techniques, especially in 3D dimensions. The aim of this project is at building a LSCM based 3D image processing technique platform and providing with a powerful support to biomedical microscopy image analysis. To reach the goal, we are undertaking intensive and extensive research in each process, from image acquisition to feature extraction. Upon the basis, we have established a frame work of the 3D image analysis and processing.Point spread function (PSF) estimation is very important in image restoration. In this project, a new method of point spread function estimation is proposed for a LSCM system. It includes the point spread function estimations of both the LSCM imaging system itself and the sample undergoing. The point spread function is called general point spread function.A new method, called NIAWT, based on wavelet transform, is proposed for intensity interpolation. The method is characterized by boosting the high frequency components, improving the interpolation accuracy, and increasing the ratio of signal to noise of the images. The experimental results show that the new method is better than the traditional linear interpolation and other wavelet transform based interpolations.In most of cases, the laser scanning confocal microscopy images, such as fluorescence stained biomedical nuclei, are blurred and have a wide range of brightness and poor intensity homogeneity. In this thesis, two methods, called stepwise threshold segmentation and contour map threshold segmentation, are proposed. By either methods, a much better segmentation of laser scanning confocal microscopy 3D biomedical images has been achieved than by the existing methods.Principal axis transform is applied to the feature extraction of 3D images and good results have been received. New methods, called 2D and 3D taking-off and landing-on methods, are proposed for distance transform. They can speed up the computation, especially in 3D distance transform. New methods of Hough transform are proposed for 2D circle and 3D sphere detections. They can reduce the memory requirement and increase the computation speed. The above methods have been applied in laser scanning confocal microscopy 3D biomedical image processing and have shown their great merits.Four improved methods are proposed for 3D object reconstruction. They are run-length based 3D object volume reconstruction, contour integration and link table method of 3D object volume reconstruction, improved marching cube 3D surface reconstruction, and improved contour following 3D surface reconstruction. The first two methods can reduce the computational time. The third method has avoid the hole problem encountered in traditional marching cube method. The last one has solved the key problems of contour pairing, surface bifurcation, and contour patching.On the basis of the above problem solving and a great number of image processing theories and techniques developing, a laser scanning confocal microscopy 3D image processing platform has been built up. By the experiments of practicalbiomedical image processing, it has been proved that the platform has theoretic significances and application merits. With precisely processing and analyzing laser scanning confocal microscopy biomedical images, providing with a great quantitative measurements, the platform can be a helpful foundation for the intensive and e...
|
PDF Full Text Request